Write Context Once, Then Ship It Like Code to Any Agent

What it does

TrustGraph is a containerized, self-hosted platform that acts as a full-stack semantic infrastructure for agentic systems. It ingests documents, automatically extracts entities and relationships, and assembles them into ontology-driven context graphs that ground agent queries in structured, explainable facts. The platform bundles storage, messaging, vector search, and inference orchestration so you can deploy locally or in Kubernetes without cobbling together a dozen external SaaS accounts.

The interesting bit

The standout idea is the “Context Core” — a portable, versioned bundle of your domain knowledge, embeddings, provenance, and retrieval policies that you treat like code: build, test, pin, promote, and roll back. It shifts context from prompt engineering to context engineering, giving you reproducible agent behavior across environments.

Key highlights

• Ships with its own data stack — Cassandra for multi-model storage, Qdrant for vectors, Pulsar for pub/sub, and Garage for object storage — so only LLM or OCR services might need external API keys
• Out-of-the-box RAG variants: DocumentRAG, GraphRAG, and OntologyRAG, plus a 3D interactive graph explorer and visual explainability DAGs
• Agent orchestration supporting ReAct, Plan-then-Execute, and Supervisor patterns, with MCP integration for tool use
• Web UI includes an ontology editor with OWL/XML round-trip import/export, a prompt editor, document ingestion workflows, and live explainability event tracking
• REST, WebSocket, Python, and CLI APIs, plus TypeScript client libraries for UI integration

Caveats

• The bundled Grafana observability dashboard ships with hardcoded admin/admin default credentials, which you’ll want to change before exposing to a network.

Verdict

Teams running private, deterministic agent workloads that need explainable retrieval and sovereign data control should look here. If you just want a quick OpenAI wrapper for a chatbot, this is overkill.